Skip to main content

Table 3 Runtimes for each clustering method

From: A systematic comparison of genome-scale clustering algorithms

  Small (3-10 genes) Medium (11-100 genes) Large (101-1000 genes)  
Clustering Method Param Time (s) Param Time (s) Param Time (s) Implementation
K-Clique Communities 0.80/03   0.80/57   0.80/48   Standalone,***
Maximal Clique 0.80 26.510 0.80 26.510 N/A N/A Standalone,***
Paraclique 0.80/01 5.120 0.80/09 0.780 0.60/09 9.050 Standalone,***
Ward (H) N/A 2.863 N/A 2.863 N/A 2.863 R 2.13.0
CAST 0.875 37.324 0.85 34.242 0.90 34.121 MeV 4.5.1
QT Clust 030 6 904.518 035 6 759.073 050 5 559.467 MeV 4.5.1
Complete (H) N/A 2.721 N/A 2.721 N/A 2.721 R 2.13.0
NNN 11 25.550 24 30.610 27 34.370 Standalone,***
K-Means 617 6 711.143 308 4 060.351 21 1 068.069 R 2.13.0
SOM 25/r 6.159 25/h 6.121 18/r 2.956 MeV 4.5.1
WGCNA 2/10 79.430 1/06 80.962 2/06 80.962 R 2.13.0
Average (H) N/A 2.452 N/A 2.452 N/A 2.452 R 2.13.0
McQuitty (H) N/A 2.445 N/A 2.445 N/A 2.445 R 2.13.0
CLICK 015 38.270 060 45.310 065 52.570 Standalone,***
  1. Parameters used to produce the best Jaccard score, and the associated runtime for the given method and parameters are displayed. Specific parameter descriptions are listed in Table 1. MeV times were reported by GUI results. Hierarchical methods use the "flashClust" package for R, which is a C++ implementation of the standard "hclust" package. Hierarchical timings do not include the time for tree cutting (which is negligible). flashClust and WGCNA packages were downloaded from the CRAN repository June 22, 2011. Versions reported refer to the version used for runtime calculation; in some cases, previous versions were used to generate clusters for scoring. r = rectangular, h = hexagonal. *A GUI-based graphical tool which is no longer maintained was used to generate clusters for Jaccard scoring while the latest R implementation was using in timing, **Total elapsed time reported by the system.